Thanks for the quick reply. That makes sense, although I am still confused why the functions prevent mincriterion from being lower than 0 as to run cforests with "testtype" as "Teststatistic" and grow maximum depth trees you need to put in qnorm(0) (= -inf) which is impossible? I will stick to other "testype"s as a work around for this.
Thanks again, Matt On 27 July 2010 16:34, Achim Zeileis <achim.zeil...@uibk.ac.at> wrote: > On Tue, 27 Jul 2010, Matthew OKane wrote: > > Hi, >> >> Could anyone help me understand how the mincriterion threshold works in >> ctree and cforest of the party package? I've seen examples which state >> that >> to satisfy the p < 0.05 condition before splitting I should use >> mincriterion >> = 0.95 while the documentation suggests I should use mincriterion = >> qnorm(0.95) which would obviously feed the function a different value. >> > > It depends on what the "testtype" is, which controls the criterion which is > computed. For "Bonferroni"/"MonteCarlo"/"Univariate" the criterion is > 1-pvalue > while for "Teststatistic" it is a type of > z-statistic > > In the former case (default in ctree), a mincriterion of 0.95 or 0.99 etc > is intuitive. In the latter case (default in cforest), a mincriterion of > qnorm(0.9) or qnorm(0.95) etc. is natural. > > hth, > Z > > Thanks in advance, >> Matthew >> >> [[alternative HTML version deleted]] >> >> ______________________________________________ >> R-help@r-project.org mailing list >> https://stat.ethz.ch/mailman/listinfo/r-help >> PLEASE do read the posting guide >> http://www.R-project.org/posting-guide.html >> and provide commented, minimal, self-contained, reproducible code. >> >> [[alternative HTML version deleted]] ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
